Hi!

I have an idea about how I'd like to publish my web server assets, integrating with go-git. I'm hoping to achieve a few things at once:

• I want to use the sendfile/2 optimization for serving assets. Go's net/http.ServeContent does this by default if you give it an open *os.File.
• I want to ensure updates are smooth, so old assets continue to be served even when a new version is published. I will be able to differentiate them because all asset requests will include the hash of the file in a query parameter.
• I want to publish new assets by pushing to a tag in a git repo.
• I'd like to do this with minimal file duplication on the host file system.

So here's the strategy I'm cooking up:

1. Use go-git to pull the latest tag to a bare clone of the asset repo with depth 1
2. ❓Have go-git unpack all the asset files into complete object files under .git/objects
3. Use go-git to resolve the tag and walk the tree, adding assets to the web server list (and parsing templates)
4. ❓When an asset request comes in, get the path to the underlying object file on disk in the .git/objects dir, open it, and serve it directly with ServeContent
5. Wire up some web hook to notify on new tags, goto 1
6. Garbage collect old tags and their objects after retiring them (TBD).

With this I'd be able to serve as many versions of files as I want and git automatically deduplicates them for me so it seems like a great plan.

My questions are:

• Is it possible to force go-git to store all the objects as separate files per object? Git objects can be hunks or packed in packfiles, etc, so I know I'm asking for something a bit unusual, but by my understanding it should be possible while remaining a valid git object dir.
• Can I get the path to such an unpacked file in the objects dir?